Are your metrics aligned with your goals? Why pLTV is different to CPA and ROAS

“We ordered a bridge to go from one side to the other. It doesn’t do that”.

- Hans Kjær, Esbjerg Director for Technology and the Environment

On 22nd September 2017, the Esbjerg, Denmark residents celebrated the inauguration of a new bridge to cross the Kongeåen Canal, the result of a major engineering project for the municipality. It’s an unexceptional story, except the bridge was a meter too short to reach over the canal. How do you set out to build a bridge for it to fail to reach the other side?

Whether we’re physicists, performance marketers, or structural engineers, accurate measurement has been, and is, essential for success. Modern marketers quantify the performance of pretty much all of their activities although, in the process, often need to make difficult compromises and tolerate imperfect metrics.

Still, it pays to be vigilant about what is being measured. It pays to remind ourselves: are we measuring the right thing? And are we measuring it correctly? It pays to know the limitations and trade-offs of our metrics and to what extent they continue to align with true objectives rather than being superficial proxies.

While optimization defines our aim, metrics help us judge progress toward that goal. When measuring the value of ad campaigns, it is common to use performance metrics like cost per acquisition (CPA), return on ad spend (ROAS), or predicted lifetime value (pLTV) per user, cohort, or campaign, plus incremental versions of metrics (comparing outcomes between groups who see and do not see your ads).

Each of the metrics emphasizes different goals (cost vs value, short vs long-term value) and provides different insights. We need to ask whether the metric we use brings us closer to our business goals and why.

• CPA describes the cost of acquiring a conversion or a user but by itself misses the larger point of how much value that conversion or user might bring in the future. To see why this is the case, let's look at a real example of a gaming company. 

  
  

  When we reviewed their historical data, we found that 10% of users contribute 94% of the revenue (see the plot below). This means a very few high-value customers drive most of the revenue. If some customers are worth more to the business then it follows we should willing to pay more to acquire them versus less valuable customers. Purely optimising towards low CPAs misses this critical insight.

  
  

  

  Following on from that, it also means they'll struggle with their advertising. Ad platforms will not automatically know these profound value differences between customers unique to this gaming company.

  
  

• ROAS is simply the revenue attributed to ads divided by the cost. However, for many digital advertisers, ROAS is calculated and attributed by an ad platform. It results from the platform's optimization windows and attribution models. Therefore, it tends to focus on immediate to short-term campaign performance, offering a snapshot of returns, e.g., modeled ROAS on day 1 or 7 

• pLTV per user describes the long-term value of individual customers in terms of predicted revenue. This kind of insight helps us decide how much to spend to reach customers because we can send this insight as a signal to the ad platform to find and attract more high-value people like them. The image below is an illustration of what your CRM could look like when filtered by pLTV  

• 

• Using pLTV as a metric to compare campaigns: We can aggregate those predictions per user to compare campaigns in terms of long-term revenue. For example, we can look in Meta Ads Manager or Google Analytics (see image below) to directly compare which campaigns drive higher predicted long-term value. Of course, we need to be mindful that these results are attributed and not causal 

• Incremental pLTV evaluates the incremental impact of long-term metrics, i.e., Is our advertising strategy driving additional long-term value on top of what we would have gotten anyway? In our Meta case study with Codeway, we can see the results of a 2-cell test comparing Churney’s pLTV strategy on Meta versus a business-as-usual strategy: 

  • 32% more incremental subscriptions compared to the usual strategy

  • 19% decrease in cost per incremental high-value customer acquisition compared to the usual strategy

  • 20% higher predicted lifetime value of newly acquired customers compared to the usual strategy

Marketers can fall short of their goals by measuring the wrong thing (vs true objective) or mismeasuring the right thing (not using causal measurement). In the case of the bridge mishap, the reason was likely due to a combination of factors stretching from planning, design, and communication, to execution. Although funnily enough, the only direct reason I found from sleuthing online was simply, ‘someone didn't measure correctly.’ Let us know on Linkedin if you find a better reason.  

We need to be constantly aware of the gap between what’s real and what’s easy to measure. Nguyen (2024) describes this in his brilliant essay on ‘The limits of data” and what happens when industries confuse the two and come to internalize metrics as core values:

 ‘... academics aim at citation rates instead of real understanding; journalists aim for numbers of clicks instead of newsworthiness.’

We can add marketers reaching for LTV measuring low CPA or high short-term ROAS rather than (incremental) higher customer lifetime value.

Using a mixture of long-term metrics together brings marketers closer to their true business goals: 

1. pLTV per user will provide unique insights into who is driving our revenue

2. Using predictions to compare campaigns on ad platforms provides early insight into campaign performance before we test 

3. Incremental pLTV gives us the confidence in knowing our strategy is causing the results